What approach can be used to effectively monitor cyclic metrics with unpredictable periods?

Question

Options

• AImplement custom anomaly detection algorithms to handle cyclic variations
• BUse LAG and LEAD analytic functions to identify cyclic patterns
• CAdjust the detector's threshold dynamically based on cyclic fluctuations
• DApply moving averages with different window sizes

• A
  70% (30)
• B
  14% (6)
• C
  5% (2)
• D
  12% (5)

Explanation

Custom anomaly detection algorithms are the right tool here because they can adapt to patterns without requiring a known, fixed period - they learn the shape of the cycle from the data itself rather than assuming regularity. Option B (LAG/LEAD) requires knowing the offset between comparable points, which is impossible when the cycle length is unpredictable. Option C (dynamic thresholds) addresses sensitivity but not the structural challenge of recognizing a cycle whose timing shifts unpredictably. Option D (moving averages with varied windows) smooths out noise but cannot reliably detect or track cycles when you don't know which window size matches the current period.

Memory tip: Think "unpredictable = can't hardcode" - any option that requires you to pre-specify a fixed offset, window, or threshold fails when the period is unknown. Only adaptive/learning approaches (anomaly detection) survive unpredictability.

No community discussion yet for this question.